Investigate optimal C++ libraries for audio DSP and music

[gogins]

The most obvious use case is writing new Csound opcodes in C++. But there may be others. Before we get going, some of my issues:

1. The whole library should be as simple as possible for what it does...
2. Which is to provide the highest precision, highest-performance audio DSP that can be used for real-time and non-real-time performance.
3. C++11.
4. Preferably, a header-file only library.
5. And of course, an open source license, though GPL is usable thanks to dual compatibility.
6. Usually the signal flow graph is crap, something closer to the state of the art would be nice.
7. Usually the developer knows either too much C++, or too little.
8. In real life, adequate performance requires processing audio sample frames in blocks.
9. Be a Linux package to integrate with the Csound build system.

• Aquila at http://aquila-dsp.org/, is open source. Last maintained in 2013 or so.
• aulib at https://github.com/aulib/aulib, GPL 3, seems to be the child of SndObj.
• CLAM at http://clam-project.org/, GPL 2, may have morphed into Essentia.
• Create Signal Library at https://svn.mat.ucsb.edu/svn/CSL/, is open source, but the repo is dead, also https://github.com/stpope/CSL.git, but that repo is empty. Wait, moved to https://github.com/stpope/CSL6, BSD license.
• eDSP at https://github.com/mohabouje/eDSP, GPL 3.
• Essentia at http://essentia.upf.edu/, AGPL 3, but it's about music informatics not synthesis.
• Faust at https://faust.grame.fr/ can be counted here, even though it is its own language, because it can generate C++ implementations of Faust signal flow graphs.
• Gamma at https://github.com/LancePutnam/Gamma, is open source.
• Jamoma at https://github.com/jamoma/JamomaCore, is open source.
• JUCE at https://github.com/WeAreROLI/JUCE.
• KFR at https://github.com/kfrlib/kfr looks interesting, GPL 3.
• Maximilian at https://github.com/micknoise/Maximilian, is open source.
• Q at https://github.com/cycfi/Q.
• SndObj at http://sndobj.sourceforge.net/, GPL 3.
• SoundTouch at https://www.surina.net/soundtouch/index.html, is open source.
• SPUCE at https://github.com/audiofilter/spuce, is open source.
• Synthesis Toolkit in C++ at https://ccrma.stanford.edu/software/stk/, is open source, and the only one I have really used. I don't like the on-disk wave tables.
• min-devkit for Max, BSD license: https://github.com/Cycling74/min-devkit

[gogins]

Probably the best path forward now is to write some new Csound opcodes and try some of these to see what works best. Key issues are (a) simplicity and (b) quality of the signal flow graph.

[gogins]

Gamma enabled me to implement MVerb without too much trouble, but there are of course things about it I do not like. I might be able to do better myself but, of course, that way lies madness.

[gogins]

Of all the things I have tried, Faust is without doubt the most capable and comprehensive. Unfortunately it is not usable as a simple library. Still it can be used like one in Csound with the Faust JIT opcodes, or used to generate C++ code for use with the Clang JIT opcodes. Other "architectures" could easily be developed based on the C++ architecture.